[dpdk-dev] [PATCH v6 2/5] lib/librte_ether: support l2 tunnel operations

[Lu, Wenzhuo]

Hi Thomas,


> -----Original Message-----
> From: Thomas Monjalon [mailto:thomas.monjalon at 6wind.com]
> Sent: Wednesday, March 9, 2016 5:28 PM
> To: Lu, Wenzhuo
> Cc: dev at dpdk.org
> Subject: Re: [dpdk-dev] [PATCH v6 2/5] lib/librte_ether: support l2 tunnel
> operations
> 
> 2016-03-09 01:15, Lu, Wenzhuo:
> > From: Thomas Monjalon [mailto:thomas.monjalon at 6wind.com]
> > > 2016-03-08 14:53, Wenzhuo Lu:
> > > > +/**
> > > > + * l2 tunnel type.
> > > > + */
> > > > +enum rte_eth_l2_tunnel_type {
> > > > +	RTE_L2_TUNNEL_TYPE_NONE = 0,
> > > > +	RTE_L2_TUNNEL_TYPE_E_TAG,
> > > > +	RTE_L2_TUNNEL_TYPE_MAX,
> > > > +};
> > >
> > > We already have rte_eth_tunnel_type.
> > Seems the tunnels in rte_eth_tunnel_type are all L3 packets. So, I want to add
> a new type for e-tag, s-tag... as they're l2 packets.
> > Do you suggest to merge it into rte_eth_tunnel_type?
> 
> Maybe you can keep the L2 prefix and add it in the same enum.
> It depends wether the rest of the API is specific to L2 or not.
OK, I'll keep the L2 prefix and merge it to the enum rte_eth_tunnel_type.

> 
> > > Why this struct is in rte_eth_ctrl.h and not used with rte_eth_dev_filter_ctrl?
> > Just want to put it together with rte_eth_tunnel_type :)
> 
> > > Why are we still adding some filtering functions after having the
> > > assertion that the new filtering API in lib/librte_ether/rte_eth_ctrl.h was
> generic enough?
> > > The filtering API v2 was a total failure.
> > > Are we going to add new functions each time a new bit of a header
> > > must be parsed by an offloaded filtering?
> > > Are we going to add new functions for each new filter of a NIC?
> >
> > Sorry, my bad. I'll try to use the existing filter API. Thanks.
> 
> OK, using the filtering API v2 is better.
> But I'm not confident it is a good API.
> If you have any concern, please discuss them. Because we need to discuss how
> to make a really generic API which fits with any filtering (flow steering) offload
> of any vendor while being descriptive enough and easy to use.
It's OK for me to use rte_eth_dev_filter_ctrl :)